A Machine Learning Approach To Analyzing Multi-Attribute Data: Development And Illustration Of The Ordeval Algorithm

Abstract

Multi-attribute models are commonly used to assess customers’ evaluative judgments. Typically, the accompanying data are analyzed using regression-based techniques that require a priori specification of the functional form, apply to only a relatively restricted set of functional forms, and implicitly assume that positive and negative changes between two data points are equal (i.e. symmetry). Building on machine learning theory our aim is to develop and illustrate an alternative approach (OrdEval algorithm) for analyzing multi-attribute data that overcomes the above-mentioned problems. Furthermore, using data from two different settings the practical application of the OrdEval algorithm is illustrated.